Detection of COVID-19 in High Resolution Computed Tomography Using Vision Transformer

Abstract

In the current pandemic, precise and early diagnose of COVID-19 patient remained a crucial task for control of the spread of the COVID-19 virus in the healthcare sector. Due to the unexpected spike in COVID-19 cases, the majority of countries have experienced scarcity and poor testing rate. Chest X-rays and CT scans have been discussed in the literature as a viable source of testing for COVID-19 disease in patients. However, manually reviewing the CT and x-ray images is time-consuming and prone to error. Taking account into these constraints and the improvements in data science, this research proposed a Vision Transformer-based deep learning pipeline for COVID-19 diagnose from CT-based imaging. Due to the scarcity of large data sets, three open-source datasets of CT scans are pooled to generate 27370 images of covid and non- covid individuals. The proposed vision transformer-based model accurately diagnoses COVID-19 from normal chest CT images with an accuracy of 98 percent. This research would assist the practitioner, radiologist and doctors in early and accurate diagnose of COVID-19.